Note : Les descriptions sont présentées dans la langue officielle dans laquelle elles ont été soumises.
CA 03076210 2020-03-17
1
Description
System and Method for Navigating Within a Track Network
Field of technology
[01] The invention relates to a system for navigating within a track
network,
comprising as system components a system central, a track maintenance
machine and communication means. The system central is set up for
administering network data which represent a model of the track network.
The track maintenance machine is suited for the treatment of track sections
of the track network, wherein the track maintenance machine comprises a
navigation device for processing navigation data derived from the network
data. The communication means are provided for data exchange between
the system central and the navigation device. In addition, the invention
relates to a method of operating the system.
Prior art
[02] Navigation within a track network is required for the operation of
track
maintenance machines. Initially, operational plans including operating
locations and work orders are compiled manually or in a partially automatized
way in a system central. Then, a transfer of operating plan lists to the track
maintenance machine takes place in order to specify the defined operating
locations as navigation targets. As a rule during this, it is assumed that a
machine operator knows the local circumstances.
[03] According to EP 1 862 593 A2, a system is known which enables an
automatized localization of a track maintenance machine within a track
network by means of a satellite-based tracking system. In this,
synchronization with characteristic data of a track line database takes place
in order to make a precise position determination.
[04] In the course of construction site planning, it may happen that an
operating
location is not clearly defined or is indicated incorrectly. For example, an
invalid reference may be noted in the operating plans. Such mistakes can
CA 03076210 2020-03-17
2
lead to a delay in work execution. In the worst case, an operation takes place
on a wrong track section.
[05] According to the prior art, operating plans are compiled on the basis
of
network data which are stored in several different databases. In this, the
databases in each case concern different installations within the track
network (for example, signalling equipment, track, catenaries, etc.). These
databases are tailored to the respective requirements of the various planning
authorities of a railway infrastructure operator, wherein, as a rule,
different
references (for example, mileage, signs for masts, sleepers, signalling device
etc.) are stored to indicate an operating location.
[06] If the track network is changed, such as, for example, when a switch
is
removed, the network data in the known databases have to be updated
manually. In practice, there are frequently problems with the consistency,
completeness, validity and precision of the available network data. In further
sequence, the deficient data maintenance leads to faulty operation planning
of the track maintenance machine.
Summary of the invention
[07] It is the object of the invention to provide an improvement over the
prior art
for a system and for a method of the type mentioned at the beginning.
[08] According to the invention, these objects are achieved by way of the
features
of independent claims 1 and 7. Advantageous further developments of the
invention become apparent from the dependent claims.
[09] In this, it is provided that the system comprises at least one movable
or
stationary carrier platform with sensors for collecting raw data which
represent characteristic information of the track network, and that a big data
framework is set up in the system central in order to evaluate the raw data
and synchronize them with the network data.
[10] In this manner, the network data are continuously adjusted to a
detected
actual state. In this, the setup of the big data framework in the system
central
enables an automatized evaluation of the collected raw data on the basis of
pre-set evaluation criteria or ¨algorithms. During this, not only conventional
databases and data analysis tools are used, but also various processes of
CA 03076210 2020-03-17
3
machine learning. In this manner, a model of the track network is constructed
on the basis of the collected raw data and continuously further developed.
Based on the model and a sensor system installed on the track maintenance
machine, an automatized real-time position determination in takes place. In
addition, work parameters can be pre-defined independent of location.
[11] In this, it is advantageous if the track maintenance machine is
designed as a
carrier platform and comprises a sensor system which collects raw data
during travel on the track network. Thus, the network data become ever more
precise solely by increasing operational use of the track maintenance
machine, so that accurate operational planning is ensured.
[12] In addition, it is useful if the system comprises as carrier platform
a
measuring vehicle or other track-bound vehicle equipped with sensors. As a
rule, a track network is travelled over at prescribed time intervals by a
measuring vehicle in order to establish an actual condition. In case of
integration into the present system, the raw data collected by means of the
measuring vehicle also serve for updating the network data. Conventional rail
vehicles may also be equipped with sensor systems for this purpose.
[13] A further improvement provides that the system comprises a flying
carrier
platform, in particular a drone equipped with sensors. Also, high-resolution
satellite images can be used as raw data in order to augment the network
data with information.
[14] In an advantageous embodiment of the invention, the network data are
stored as a graph with track objects as nodes and with relations between the
track objects as edges. In this way, the network data can be administered in
a simple manner. In particular, the synchronization with the evaluated raw
data is facilitated by suitable algorithms.
[15] In this, it is favourable if characteristic data patterns are
specified for a track
object. Basis for this are significant features common to every object in a
certain class (for example, sleepers, rail fastening means, light signals,
etc.).
As a consequence, depending on the sensors used, corresponding data
patterns are found in the raw data which ensure an efficient object
allocation.
[16] The method, according to the invention, for operating one of the afore-
mentioned systems provides that the raw data are collected by means of the
CA 03076210 2020-03-17
4
sensors, that the raw data are transmitted to the system central, that object
data are generated from the raw data by means of identification algorithms,
and that the network data are synchronized with the object data in order to
update the network data.
[17] In this, the big data framework set up in the system central serves
for
automatic evaluation of the collected raw data. With this method, an
automatized adaptation of the network data takes place as soon as
corresponding raw data are collected by means of the sensors. Thus,
problems based on deficient data maintenance are precluded.
[18] In a favourable further development of the method, after an update of
the
network data has taken place, all the updated data or part of the updated
data are transferred to the navigation device of the track maintenance
machine. If the system encompasses several track maintenance machines, a
corresponding data transfer takes place to all machines. In this way, the
navigation to the next operating location is always based on updated network
data.
[19] An advantageous embodiment of the method provides that probability
values
or probability functions are assigned to the object data in each case in
dependence on the sensors used and/or the carrier platform used and/or the
identification algorithms used. In this manner, an identification precision in
relation to the data stock or the track network is ascertained. Thus, a
classification of the determined objects takes place with respect to their
information content for the track network or for the already existing track
network information.
[20] The machine learning in the scope of the big data framework enables
the
continuous expansion and adaptation of the recognizable objects on the
basis of new raw data. For example, definitions or algorithms deposited in an
object register are updated with new data.
[21] In this, it is favourable if an update of the network data by new
object data is
carried out in dependence on the assigned probability values or probability
functions. The network data are thus brought up-to-date in that new object
data are added on the basis of the evaluated information content.
CA 03076210 2020-03-17
[22] A further improvement provides that the object data are organized on
the
basis of a detected motion pattern of the carrier platform, so that track
objects represented by the object data, strung together as an object chain,
are supplied for the synchronization with the network data stored as a graph.
This simplifies the synchronisation because, due to the detected motion
pattern, a logical sequencing of several track objects takes place. Thus, the
significance of an object chain with complex structure is derived from the
significance of the individual track objects.
[23] Advantageously in this, the object chain is subdivided into segments,
wherein
a segment is synchronized with the graph on the basis of distinctive track
objects. This method step likewise optimizes the synchronization of the
network data with the object data.
[24] A simple to carry out synchronization of a segment with a partial
graph
provides that an extent of agreement is specified, and that the partial graph
is
replaced by the segment if the extent of agreement exceeds a pre-set
minimum extent.
[25] The fault tolerance of the method is heightened if, when synchronizing
a
segment to a partial graph, a non-verifiable track object remains as node of
the partial graph until a pre-set number of failed verifications has been
reached. In this manner, sensor- or transmission defects have no influence
on the quality of the network data.
[26] For a navigation procedure, it is advantageous if surrounding track
objects
are recorded by means of sensors arranged on the track maintenance
machine, and if a current position of the track maintenance machine is
determined by synchronization of the recorded track objects with the network
data. In this, changes of the track network are automatically taken into
account during navigation of the track maintenance machine.
Brief description of the drawings
[27] The invention will be described below by way of example with reference
to
the accompanying drawings. There is shown in a schematic manner in:
Fig. 1 track maintenance machine
Fig. 2 track network
CA 03076210 2020-03-17
, 1
6
Fig. 3 system layout
Fig. 4 structure of track sections
Fig. 5 method sequence
Description of the embodiments
[28] A track maintenance machine 1 shown in Fig. 1, which is to be
navigated to a
work assignment, is a component of the system according to the invention.
This track maintenance machine 1 comprises working units 2 for treatment of
a track section 3 of a track network 4. The track network 4 includes various
objects such as, for example, tracks 5, switches 6, crossings 7, masts 8,
tunnels 9, stations 10, underpasses 11, level crossings 12 or balises 13, as
shown by example in Fig. 2.
[29] The track maintenance machine 1 is further equipped with various
sensors or
sensor systems 14 in order to record the surroundings of the track 5 travelled
upon, and the current position. These are, for example, a camera 15, a
positioning system 16, a clearance gauge scanner 17 or a rail scanner 18. In
this manner, the track maintenance machine 1 functions as a carrier platform
for the sensors or sensor systems 14.
[30] In order to get to a track section 3 to be worked on, the track
maintenance
machine 1 has a navigation device 19. This is configured as a computing-
and controlling unit and serves for navigation within the track network 4
which
is represented by network data. By means of the navigation device 19,
navigation data derived from the network data are processed and
synchronized with sensor data in order to determine the current position of
the track maintenance machine 1.
[31] One object of the present invention is to continuously update the
network
data in an automatized way. To that end, at raw data representing
characteristic information of the track network 4 are first collected by means
of the sensors or sensor systems 14. In further sequence, the raw data are
evaluated and synchronized with the network data administered in a system
central 20. From this data synchronization it is possible to draw conclusions
with regard to the condition of the track network 4 or individual track
sections
CA 03076210 2020-03-17
, I
7
3. For example, a frequent change of collected position data allows
conclusions as to an unstable track position.
[32] Besides the track maintenance machine 1, other carrier platforms can
be
used for collecting the raw data, for example a measuring vehicle 21, another
rail vehicle 22 or a flying carrier platform 23. Infrastructure facilities
equipped
with sensors 14 can be used as a stationary carrier platform 24. This might
be, for example, a mast 8 with a camera 15 fastened to it which observes a
track section 3. Also, fibre optic cables installed adjacent to the track 5
can
be used as sensors 14. The basis for this is the so-called Distributed
Acoustic Sensing (DAS) in which laser impulses are sent through fibre optic
cables in order to thus register in real time along a track section 3 sound
signals and activities which can be derived therefrom. By means of such
stationary carrier platforms 24, raw data of an observed track section 3 are
collected over time. Aside from detecting object changes, these raw data can
also be used for position verification of moving carrier platforms 1, 21, 22,
23.
[33] The measuring vehicle 21 is equipped, for example, with a GNSS
receiver
25, a clearance gauge scanner 17 and a rail scanner 18. The other rail
vehicle 22 includes a GNSS receiver 25, and the flying carrier platform 23
comprises a camera 15 or other devices for recording aerial views. By means
of all these sensors or sensor systems 14, various raw data are collected and
supplied for evaluation. In this, depending on data volume and available
computing power, the raw data are either pre-processed on the carrier
platform 1, 21, 22, 23, 24 or transmitted directly to the system central 20.
[34] As can be seen in the illustrative system layout in Fig. 3,
communication
means 26 are provided as additional system components. These
communication means 26 serve for data exchange between the system
central 20, the sensor systems 14 and the navigation device 19. These
means are, for example, devices for wireless communication via a mobile
radio network. In a very simple embodiment of the system according to the
invention, the system central 20 is accommodated in the track maintenance
machine 1, so that there is a self-sufficient system for navigation and
network
data updating. In this, the communication means 26 may be elements of an
installed bus system.
CA 03076210 2020-03-17
8
[35] In the system central 20, a big data network 27 is installed. This
also
supports various machine learning algorithms besides conventional
databases 28 and data analysis tools. Examples for this are noSQL or
Hadoop. In this manner, the system central 20 serves for collecting, storing
and processing the data.
[36] From data points .774,s(t), the sensor systems 14 generated a data
tensor
Sk(t) with an arbitrary dimension k at a point in time t:
Sk(t) = (Mi(t),
wherein i = 1, k are Mi r x s ¨ are matrixes, there fore
m,1(t) = = = m(t)
Mi(t) =
,i(t) = = = ml.,s(t)
Sensors 14 arranged on moving carrier platforms 1, 21, 22, 23 deliver in
particular data points 774,s(t) with spatial information. In stationary
carrier
platforms 24, however, the data points recorded by means of sensors 14
display especially temporal changes.
[37] From characteristic features in the data points inks(t) of the sensor
systems
14, virtually indexed objects 29 (objekti) are compiled as object data. These
represent track objects 4-13 which are recognizable with stochastic
reliability
in the track environment and can serve as reference for navigation.
Specifically, the objects 29 are characterized by significantly reproducible
patterns. In an object register, the definitions (or algorithms) of the
objects 29
are updated by new data.
[38] During this, a probability is calculated for each object
characteristic or a
probability function Px is assigned:
CA 03076210 2020-03-17
9
objekt id Pict \
time t Pt
coordinates x,y,z Px,y,z
offset Ax, Ay, Az PAxAyAz
objekt,i(t) = value At PAt
next objekt 01+1 Poi+i
PAtoi+i
data points M1 pmi
,
\ data points Mx pitix
Aside from the metadata resulting from the object verification, the data
points
in particular describe a current state of the virtually indexed object 29.
[39] The respective probability function P is dependent on the type of
sensor or
sensor system 14, the type of carrier platform 1, 21, 22, 23, 24 and the
algorithms in the object register. For example, a probability function with
little
scattering is prescribed for newer sensor systems 14. For older sensors 14
with less precision, however, a greater scattering is prescribed. A
recognition
probability derived from stored observation processes takes into account the
number of objects 29 recognized so far. In this way, a degree of
unambiguousness of an object 29 in relation to a data stock stored in the
system central 20 is established. Accordingly, a classification of the objects
29 by means of the evaluated information content takes place with respect to
the total track network and the information therein.
[40] In addition, the evaluation method is continuously improved in that
the
recognizable objects 29 are continuously expanded and adapted by means
of machine learning or manual learning. In this, the autonomous machine
learning within the big data framework is based on those data which are
newly collected in the system central 20.
[41] As shown in Fig. 4, virtually indexed objects 29 are linked relative
to one
another according to a motion pattern of the carrier platform 1, 21, 22. Thus,
in the case of track-bound carrier platforms 1, 21, 22 strung-together
virtually
indexed objects 29 are generated, wherein the resulting object chain 30 can
be regarded as a semantic chain Ktn:
Ktn = (objekt,i(t,), , objekt,,i(t,"õ)),
CA 03076210 2020-03-17
wherein tn is a starting time and tn+m is an ending time of a run of the
carrier
platform 1, 21, 22.
[42] Within this chain, distinctive objects 29 (for example, object data of
a switch
6) are specified as discretization points for subdivision into segments 31.
These can be determined dynamically, from the probability of recognition or
from the relevancy in the network (degree of unambiguousness). In Fig. 4 for
example, the strung-together objects 29 represent, from left to right, a
switch
6, a balise 13, a track 5, a mast 8, a tunnel portal, a tunnel 9, a tunnel
portal,
a switch 6, a mast 8, a balise 13, two masts 8, and a switch 6.
[43] New object chains 30 are synchronized with the network data, i.e. the
model
31 of the track network 4. In this, the network data are stored as a graph
N(t), wherein ta is a time of updating of the particular object 29:
N(ta) = (objekty(t), edge(0)
N(tnew) = objektyi(t) ¨0 N(t01)
[44] Segments 32 of the object chain 30 are synchronized with the model 31
of
the track network 4 (mapped) via distinctive objects 29. If a segment 29 and
a partial graph coincide with a high probability, then the virtually indexed
objects 29 contained in the segment 29 are transferred to the graph. In this
manner, the characteristics of the particular virtually indexed object 29 are
used for updating the characteristics of the model 31 (update of the network
data) while taking into account the associated probability functions P. In
this,
with growing collection of raw data, the reliability and precision of the
network
data increases:
objekt id Pid
update ta Pt,
coordinates x,y,zPx,y,z next objekt Ov+1
N(ta) = value d Pd distance next objekt x,y,z Px,y,z
data points Mi. Pm,
\\ data points Mx Pmx
CA 03076210 2020-03-17
4 d
11
,
Data points collected by means of stationary carrier platforms 24 are also
synchronized in a corresponding manner with the network data, wherein here
the information about temporal changes is paramount.
[45] During an updating process, it may happen that, based on sensor faults
or
obstructions during object detection, individual objects 29 cannot be
verified.
Then it is practical if these remain existent in the network data until -
during a
new collecting procedure 33 - a falsification takes place, or a verification
fails
to happen several times.
[46] Shown Fig. 5 is the method sequence in an overview. In this, the
method is
divided roughly into the following method steps:
- input 34 of the various carrier platforms 1, 21, 22, 23, 24
and the sensor
systems 14
- signal analysis 35 for object indexing with statistical values
- repeated collecting procedures 33 (observations) for
verification or
falsification of the object indexing
- mapping 36 of various collection types or observation
perspectives
- output 37 of the track network model 31 (updated network data)
- navigation procedure 38 of a track maintenance machine 1
[47] In an initial phase, the system forms an initial model 31 for the
track network
4 on the basis of the pre-set algorithms. To that end, for example, aerial
views, raw data of a measuring vehicle 21, raw data of the track maintenance
machine 1 and raw data of another vehicle 22 are evaluated. In Fig. 5, the
respective signal analysis 35 initially results in different objects 29 or
object
classes which can be associated with individual track objects 5-13. In aerial
views, these are, for example, tracks 5, switches 6, tunnels 9, masts 8,
stations 10, etc. The raw data of the track-bound carrier platforms 1, 22, 23
can be associated with the track objects tracks 5, switches 6, tunnels 9, etc.
[48] In the course of the repeated collecting procedures 33, the individual
objects
29 are verified or falsified with regard to their relation to one another. In
the
case of the track-bound carrier platforms 1, 22, 23, object chains 30 are
formed which depict a track section 3 travelled upon. A merging of these
evaluation results by means of mapping 36 results at last in the model 31 of
the recorded track network 4.
CA 03076210 2020-03-17
12
[49] For reliably carrying out a navigation procedure 38, the network data
are
transmitted from the system central 20 to the navigation device 19 of the
track maintenance machine 1 after a data update has taken place or at
prescribed time intervals. In this, it is favourable if, via the desired
confidence, it is parameterized which objects 29 contained in the network
data are transmitted for navigation to the track maintenance machine 1.
[50] During the navigation procedure 38 itself, objects or object data 29
are
synchronized with those track objects 5-13 which are currently detected in
the surroundings of the track maintenance machine 1 by means of the
sensors or sensor systems 14. In this way, detected track objects 5-13 serve
as reference for position determination. Additionally, results of a track
gauge
measurement can be used for more precise positioning on the track 5. In this,
the detected progression of the track gauge forms further data points of those
objects 29 which represent the corresponding track 5. In an extension, it is
also possible to use detectable characteristic features of sleepers 39 or
rails
40 (markings, material characteristics, etc.).
[51] Additionally, the raw data collected during the navigation process 38
serve as
new data input for updating the network data. With the present system,
changes in the track network 4 due to reconstruction or malfunctions are
automatically taken into account for subsequent navigation procedures 38.
Cognition takes place cause- or time-related automatically in dependence on
the speed of migration of the track network 4.
Favourably, recognized and anticipated objects 29 in the current
surroundings of the track maintenance machine 1 are displayed to a machine
operator 41 for orientation. Work instructions can additionally be included in
this display. Also, work parameters can be prescribed to the machine
operator 41 in a location-dependent way or transmitted directly to a working
unit 2. In this way, an automatized local adaptation of work parameters takes
place, thus enabling an optimized operation of the track maintenance
machine 1. In a tamping machine, these are, for example, lifting- and lining
values as well as time stipulations for the tamping cycles. In other track
maintenance machines 1, work parameters like ballast demand, spoil
quantities of old ballast can be adjusted dependent on location.
